Headliner Clustering
Column
Commentary
Coachella has had a wide array of headliners over the past six years, ranging over diverse genres such as KPOP, rap, psychedelic rock, classic rock, R&B, and more. While each of these headliners earned their spot by being a legend and paving their own way, we can still do some grouping amongst them. The dendrogram to the right is a visualization of a clustering technique applied to a playlist I created which simply contains each headliner for the past six years represented by their #1 song. Of course, this isn’t the best method, since perhaps the artist’s #1 song is not very representative of their style, but it should be good enough to get some general groupings.
The clustering was calculated based on pitch and timbre features of the tracks, and the dendrogram was created using complete linkage. Looking on the lowest level, we can perhaps start to make out some subgenres and make sense of the clusters.
• The Less I Know The Better and Sweet Child O’ Mine were grouped together, but interestingly enough they were separated pretty distinctly from Creep by Radiohead. Sweet Child O’ Mine is classic rock at its finest, and TLIKTB by Tame Impala is in a similar world, in psychedelic rock.
• Really interestingly, perhaps by coincidence, both of The Weeknds songs in this corpus ended up perfectly together. This is interesting because Moth To A Flame is a Swedish House Mafia song with a Weeknd feature. Perhaps his style is influential enough on that track. This is where my assumption might hurt this visualization — we can’t take this to mean that Swedish House Mafia and The Weeknd should always be grouped together.
• Redbone, 7 rings, and lovely are, in my opinion, the most radio/mainstream pop songs on the playlist, and they ended up together. Perhaps the 4 chord, standard rhythm patterns signaled that to our model.
Column
Dendrogram
Dendrogram with Track Names
Introduction
An online portfolio for Computational Musicology at the University of Amsterdam (Spring 2023) by Owen Halpert
I grew up with Coachella — as a child I was surrounded by its influence in the Los Angeles region, in high school I was lucky enough to attend all four years, and in college I road tripped with my fellow students for the 2022 lineup.
I owe a lot of my passion for music and my broad taste in genres to attending Coachella. With an extremely diverse lineup, attendees can walk from one stage to the next and instantly be transported between wildly different musical experiences.
Even with this diversity, every year there are critics of the released Coachella lineup, as seen through the Reddit lineup announcement comments:
Yea….this isn’t the festival I went to in 2010😔 -bigbrett666
Awesome if you like pop and edm but sucks if you like live music. -dpcpv75
This Coachella lineup is MID. No catering to the hip hop and rap fans. SMH debating selling my box -kmart2588
Some people say too much electronic, some say too little. Some say the artists are too new and unrecognizable, some complain about too many throwback artists and want more relevance.
While there is always some validity in each of these criticisms, for my project I’d like to further analyze the trends of Coachella artist selection. To do so, I will use musicological metrics to show variance in genre, danceability over the years, and changing levels of emotional “valence” (A Spotify measure from 0.0 to 1.0 describing the “musical positiveness” conveyed by a track). I will look at artists’ live performances at Coachella, analyze melodic aspects of different tracks, and more.
The Spotify tracks I will use to analyze this progression will be yearly lineup playlists. I can then analyze and compare years as a whole. These playlists were assembled by user Alex Rodriguez, who explained that the playlists are made up of “Top 5 songs per artist, attempt to spread across their notable releases and shorter edits where possible.”
Shortcomings with this methodology might be that the selected tracks for each artist may not be representative of their true genre, that the playlists may not always be fully inclusive of all artists on the lineup, or that they are simply so large that meaningful comparisons by some metrics will be difficult to make. Nonetheless, we can use these as a starting point to focus in on some trends within instead of trying to analyze the years as a whole.
To give a sense of the breadth of these playlists, we can examine the outliers. Looking at the 2023 lineup playlist, an extremely atypical track is BTSTU by Jai Paul. Paul was added to the lineup as an artist with a cult following who is extremely secretive and has never performed live. He is as elusive, if not more, than headliner Frank Ocean. Conversely, we see One Kiss by Calvin Harris, which we can say is much more typical, given that Harris has performed at Coachella 8 times and is one of the most mainstream EDM artists.
Recent Year Overviews
Column
Commentary
The below charts give a high-level broad overview of major audio features for tracks in the 2022 and 2023 playlists. Specifically, captured are the valence, energy, danceability (size of dots, mouse over for values) and mode. With the size of these datasets it is difficult to make broad comparisons, but some observations could be as follows:
• While 2022 shows most of the high valence songs having a major key, 2023 is more of a mix, with some of the highest valence songs being in a minor key.
• There are slightly fewer high valence, low energy tracks in 2022 than there are in 2023.
• Similar to the first point, upon isolating the minor key tracks (this can be done by clicking the legend) they seem to be on average higher energy and valence than the minor key tracks in 2022.
In the next page, we will further examine danceability.
Column
2022
2023
Deeper into Danceability
Commentary
Over the past seven years, with the rise of EDM in pop culture and in Coachella’s lineups, one may ask if the festival is becoming more dancey. What we see below is the average tempi and danceability scores (a Spotify quantification) from each year. This is no easy task computationally speaking, in fact, the graph below involves over 6,000 data points, around 1000 songs curated from each year’s lineup by Alex Rodriguez.
We can see the average tempo mostly decreasing, and the average danceability mostly increasing. This is interesting, because normally you would assume these metrics go hand in hand. In reality, Spotify doesn’t explicitly give their calculation methodology for Danceability. They simply say,
“Danceability describes how suitable a track is for dancing based on a combination of musical elements including tempo, rhythm stability, beat strength, and overall regularity. A value of 0.0 is least danceable and 1.0 is most danceable.”
So, if Danceability were to increase with tempo decreasing, maybe there has been a rise in these other categories of rhythm stability, beat strength, and overall regularity. If you look at this graphic by Billboard, you can see a sharp increase in house/dance music and steady prevalence of electronic, which may explain the uptick in danceability.
Danceability
Beyoncé 2019
“Formation” Live Chromagram
An important part of Coachella history (and history in general) is Beyoncé’s “Homecoming” live album, which is her headline performance from Coachella 2019. This epic show was almost two hours long, and had the production quality of a Super Bowl halftime show. A Netflix documentary was made to immortalize the performance, something that is extremely rare for Coachella artists. Here, we will look at chromagrams for one of the songs she performed, “Formation”, and use dynamic time warping to observe how the life performance differs from the album version.
“Formation” Recording Chromagram
TODO: ADD ANALYSIS
Dynamic Time Warping
Unfortunately, the DTW is too large of a computation to render in
ggplotly, so an image is included in its place.
Proper alignment in a DTW visualization will look like a diagonal line, which we can very, very faintly see at certain time windows in the graph. This means that at some points in the performance, it was very well matched up to the original recording, and vice versa. However, with the differences in timing between the two performances, it may be hard to accurately compare them.
Frank vs. BTS
Cepstrograms
For this week’s homework we add a chroma and timbre analysis of songs by two of the three 2023 headliners to see how they may differ. Here is an analysis of “Futura Free” by Frank Ocean, and “Butter” by BTS, through the lens of cepstrograms and self-similarity matrices.
Here in the cepstrograms, we can see how Butter by BTS is split into about 5 sections. The intro has some different elements than the rest of the song, and we can discover upon listening that the intro is the one part in the song that has very little instrumental, mainly just voice and drums. Once the instrumental comes in, the first few MFCC’s become more present.
In the cepstrogram of Futura Free by Frank Ocean, there is a lot going on, but maybe most interesting is around 280-315 seconds, when almost all the MFCC’s are below 0.5 except C02. Upon listening, this is around 2:38 when the instrumental swells up, there is distorted noise, then we hear a focus on Frank’s voice over very light, sparse piano chords. The section from 315 onwards is an audio collage of different interviews and mainly speaking.
“Butter” Self-Similarity Matrices
Now onto the self-similarity matrices. The purpose of comparing these two headliners is to show that the headliners encompass a range of genres. BTS is more standard pop music, with most of their music sounding formulaic in the genre of pop. Frank Ocean is more experimental and complex in his instrumentals. Using a self-similarity matrix is a great way to compare these overall complexities.
For BTS, we see a distinct checkerboard pattern. Here, we can use that to say there is a lot of homogeneity in the song. There are blocks in the song to represent different sections, but overall there is a lot of similarity between early and late sections in the song, chroma-wise. In the timbre section, the brighter crosses represent novelty, or something new in the song. We don’t see much of this, and the timbral analysis shows that most of the song has fairly similar instrumentation, apart from the initial 25 or so seconds.